Path: utzoo!attcan!uunet!mcsun!ukc!edcastle!aipna!cam From: cam@aipna.ed.ac.uk (Chris Malcolm) Newsgroups: comp.ai Subject: Re: What actually is AI? Message-ID: <3029@aipna.ed.ac.uk> Date: 15 Sep 90 16:15:02 GMT References: <90241.112651F0O@psuvm.psu.edu> <90243.142616F0O@psuvm.psu.edu> <6560@uklirb.informatik.uni-kl.de> <2982@aipna.ed.ac.uk> <1990Sep7.203744.4326@cs.rochester.edu> Reply-To: cam@aipna.ed.ac.uk (Chris Malcolm) Organization: Dept of AI, Edinburgh University, UK. Lines: 85 In article <1990Sep7.203744.4326@cs.rochester.edu> yamauchi@cs.rochester.edu (Brian Yamauchi) writes: >In article <2982@aipna.ed.ac.uk>, cam@aipna.ed.ac.uk (Chris Malcolm) writes: >> As Brian Yamauchi has pointed out, it may well be rather silly expecting >> a *program* to display intelligence. >Well... that's not exactly what I meant. Although I was defining AI as >a research field, I *do* think it is possible for systems (autonomous >robots, for example) to display intelligence -- at least at the level of >very primitive animals (e.g. insects). I suspect we do agree: the distinction I was making was between a *program* and an *autonomous system* such as a robot. Consider a robot that we agree is intelligent, and has a computer for a brain. The intelligence of the robot does not necessarily imply that there is an intelligent program running in the computer (robot's brain). Consequently, if you are trying to build an intelligent robot, and your strategy is to first of all build a stupid robot, expecting later to be able to add intelligence in the form of a driving computer running an intelligent program, then you may well be giving yourself an unnecessarily difficult -- or even impossible -- task. Yet this is the strategy implied by many AI research programmes. And I thought I had seen Brain Yamauchi making similar sorts of distinction between programs (which run in computers) and systems (which may contain programs as components, and which run in the physical world). >> In other words, AI is a label properly applied at the moment to research >> activity rather than artefacts (such as computer programs or robots) -- >> because we currently can't make any (artificially) intelligent >> artefacts. So by this definition there is nothing in a *program* which >> makes it AI or not; what makes a program AI or not is whether it taught >> its author anything original and useful about the nature of >> intelligence. >I think this is a valid way to judge the usefulness of a research >program, but I also think it is reasonable to say that certain >"intelligent systems" do display certain limited forms of intelligence. >Some display low-level animal-like intelligence (behavior-based robots, >etc.) while others display very narrow and very brittle slices of human >intelligence (expert systems, chess-playing programs, etc). Quite so, and behaviour-based robots don't necessarily owe their "intelligence" to computer programs. In fact, I hope -- just to make the point -- to be able to show (very!) limited "intelligence" in a robot without any computational facilities. "Intelligent" is an adjective that describes a kind of purposeful and adaptable behaviour by a creature (autonomous system) going about its business in its environment. I suspect that analysis of the meaning of "intelligent" will inevitably involve reference to the physical composition and structure of the creature, its environment, its purposes, and how these all interact (sensors and effectors); and that describing the interaction cannot be done except by a multi-level description, in which lower levels create the terms in which higher levels are described (like -- but more complex then -- multiple levels of interpretation in computer languages). What the brain (or computer) does may well be an essentail component of intelligence, but IMHO it cannot possibly be intelligent on its own, any more than an internal combustion engine can fly or a floating-point processor see. What misleads many people is the supposition that intelligence is what clever well-educated people (such as professors) have more of than those who did badly at school. What professors can do better is formal abstract reasoning, as displayed in argument, maths, expert diagnosis, etc.. It is easy to make computers do these things that professors are good at, MUCH easier than what everybody can do without thinking, such as walk, catch balls, and observe the moving world spread out before them in full three-dimensional splendour. Although we have implemented lots of "professorish" AI systems based on various kinds of reasoning, we hesitate to call them "intelligent" because they have no common sense or general knowledge, demonstrating hilarious stupidities as soon as one steps beyond their narrow domains of competence. And implementing a system with common-sense -- which we all have -- is going to be many orders of magnitude harder than implementing the specialised abilities of this or that professor. After all, why should our educational and social scales of "cleverness" ranking be a good guide to the functional underpinnings of intelligent behaviour? IMHO trying to attack artificial intelligence by making systems do what we consider to be "clever" is about as silly as trying to make systems with good table manners. -- Chris Malcolm cam@uk.ac.ed.aipna 031 667 1011 x2550 Department of Artificial Intelligence, Edinburgh University 5 Forrest Hill, Edinburgh, EH1 2QL, UK